THREE-DIMENSIONAL SIMULATIONS OF STANDING ACCRETION SHOCK INSTABILITY

IN CORE-COLLAPSE SUPERNOVAE
Wakana Iwakami,
1
Kei Kotake,
2, 3
Naofumi Ohnishi,
1, 4
Shoichi Yamada,
5, 6
and Keisuke Sawada
1
Received 2007 October 10; accepted 2008 January 21
ABSTRACT
We have studied nonaxisymmetric standing accretion shock instabilities, or SASI, using three-dimensional (3D)
hydrodynamical simulations. This is an extension of our previous study of axisymmetric SASI. We have prepared a
spherically symmetric and steady accretion ow through a standing shock wave onto a protoYneutron star, taking into
account a realistic equation of state and neutrino heating and cooling. This unperturbed model is meant to represent ap-
proximately the typical postbounce phase of core-collapse supernovae. We then added a small perturbation ($1%) to
the radial velocity and computed the ensuing evolutions. Both axisymmetric and nonaxisymmetric perturbations have
been imposed. We have applied mode analysis to the nonspherical deformation of the shock surface, using spherical
harmonics. We have found that (1) the growth rates of SASI are degenerate with respect to the azimuthal index mof the
spherical harmonics Y
m
l
, just as expected for a spherically symmetric background; (2) nonlinear mode couplings pro-
duce only m 0 modes for axisymmetric perturbations, whereas m 6 0 modes are also generated in the nonaxisym-
metric cases, according to the selection rule for quadratic couplings; (3) the nonlinear saturation level of each mode is
lower in general for 3Dthan for 2D, because a larger number of modes contribute to turbulence in 3D; (4) low-l modes
are dominant in the nonlinear phase; (5) equipartition is nearly established among different m modes in the nonlinear
phase; (6) spectra with respect to l obey power laws with a slope slightly steeper for 3D; and (7) although these features
are common to the models with and without a shock revival at the end of the simulation, the dominance of low- l modes
is more remarkable in the models with a shock revival.
Subject headinqg s: hydrodynamics instabilities neutrinos supernovae: general
Online material: color gures
1. INTRODUCTION
Many efforts have been made to construct multidimensional
models of core-collapse supernovae (see Woosley & Janka 2005,
Kotake et al. 2006 for reviews), prompted by accumulated obser-
vational evidence that core-collapse supernovae are commonly
globally aspherical (Wang et al. 1996, 2001, 2002). Various mech-
anisms for producing the asymmetry have been discussed, includ-
ing convection (e.g., Herant et al. 1994; Burrows et al. 1995; Janka
& Mueller 1996), magnetic eld and rapid rotation (see, e.g.,
Kotake et al. 2006 for collective references), standing (stationary,
spherical) accretion shock instability, or SASI ( Blondin et al.
2003; Scheck et al. 2004; Blondin & Mezzacappa 2006; Ohnishi
et al. 2006, 2007; Foglizzo et al. 2006), and q-mode oscillations
of protoYneutron stars ( Burrows et al. 2006). Most of these,
however, have been investigated only with two-dimensional
(2D) simulations.
Recently, a 3D study of SASI was reported by Blondin &
Mezzacappa (2007). In 2D, the shock deformation by SASI is
described with Legendre polynomials P
l
(0), or spherical har-
monics Y
m
l
(0. c) with m 0. Various numerical simulations have
demonstrated unequivocally that the l 1 mode is dominant and
that a bipolar sloshing of the standing shock wave occurs, with
pulsational strong expansions and contractions along the sym-
metry axis ( Blondin et al. 2003; Scheck et al. 2004; Blondin &
Mezzacappa 2006; Ohnishi et al. 2006, 2007). In 3D, on the other
hand, Blondin & Mezzacappa (2007) perturbed a nonrotating
accretion ow azimuthally and observed the dominance of a
nonaxisymmetric mode with l 1, m 1, which produces a
single-armed spiral in the later nonlinear phase. They claimed
that this spiral SASI generates a rotational ow in the accre-
tion (see also Blondin & Shaw [2007] for 2D computations in
the equatorial plane), and that it may be an origin of pulsar spin.
However, many questions regarding 3D SASI still remain to
be answered. Here we are particularly interested in investigating
how the growth of SASI differs between 3D and 2D; in particu-
lar, the change in the saturation properties should be made clear.
Another question is the generation of rotation in the accretion
ow by SASI ( Blondin & Mezzacappa 2007); its efciency and
possible correlation with the net linear momentum should be
studied more in detail and will be the subject of a future paper
( W. Iwakami et al. 2008, in preparation). We also note that neu-
trino heating and cooling were neglected and the ow was as-
sumed to be isentropic in Blondin & Mezzacappa (2007), but
the implementation of this physics is helpful in considering the
implications for the shock revival and nucleosynthesis (Kifonidis
et al. 2006).
In this paper, we have performed 3D hydrodynamic simula-
tions, employing a realistic equation of state (Shen et al. 1998) and
taking into account the heating and cooling of matter via neutrino
emission and absorption on nucleons, as done in our 2D studies
(Ohnishi et al. 2006, 2007). The inclusion of neutrino heating al-
lows us to discuss how the critical luminosity for SASI-triggered
A
1
Department of Aerospace Engineering, Tohoku University, 6-6-01 Aramaki-
Aza-Aoba, Aoba-ku, Sendai 980-8579, Japan; iwakami@rhd.mech.tohoku.ac.jp.
2
Division of Theoretical Astronomy, National Astronomical Observatory of
Japan, 2-21-1, Osawa, Mitaka, Tokyo 181- 8588, Japan.
3
Max-Planck-Institut fur Astrophysik, Karl-Schwarzshild-Strasse 1, D-85741
Garching, Germany.
4
Center for Research Strategy and Support, Tohoku University, 6- 6- 01
Aramaki-Aza-Aoba, Aoba-ku, Sendai 980- 8579, Japan.
5
Science and Engineering, Waseda University, 3-4-1 Okubo, Shinjuku, Tokyo
169-8555, Japan.
6
Advanced Research Institute for Science and Engineering, Waseda Univer-
sity, 3- 4-1 Okubo, Shinjuku, Tokyo 169- 8555, Japan.
1207
The Astrophysical Journal, 678:1207Y1222, 2008 May 10
# 2008. The American Astronomical Society. All rights reserved. Printed in U.S.A.
explosion could be changed in 3Dfromthose in 2D. To answer the
questions raised above, we vary the initial perturbations as well as
the neutrino luminosity, and compare the growth of SASI be-
tween 2Dand 3Din detail, conducting a mode analysis for both
the linear phase and the nonlinear saturation phase.
The plan of this paper is as follows. In x 2, we describe the
models and numerical methods, showthe main numerical results
in x 3, and conclude the paper in x 4.
2. NUMERICAL MODELS
The numerical methods we employ are based on the code
ZEUS-MP/2 ( Hayes et al. 2006), which is a computational uid
dynamics code for the simulation of astrophysical phenomena,
parallelized by the MPI (message-passing) library. The ZEUS-
MP/2 code employs Eulerian hydrodynamics algorithms based
on the nite-difference method with a staggered mesh. In this study,
we have modied the original code substantially according to the
prescriptions in our preceding 2Dsimulations (Ohnishi et al. 2006,
2007).
We consider spherical coordinates (r. 0. c) with the origin at
the center of the protoYneutron star. The basic evolution equations
describing accretion ows of matter attracted by a protoYneutron
star and irradiated by neutrinos emitted from the protoYneutron
star can be written as
d,
dt
,:= v 0. 1
,
dv
dt
:P ,: := Q. 2
,
d
dt
e
,
_ _
P:= v Q
E
Q : :v. 3
dY
e
dt
Q
N
. 4

GM
in
r
. 5
where ,, v, e, P, Y
e
, and are the density, velocity, internal en-
ergy, pressure, electron fraction, and gravitational potential, re-
spectively, and G is the gravitational constant. The self-gravity
of matter in the accretion owis ignored. Here Qis the articial
viscous tensor, and Q
E
and Q
N
represent the heating/cooling and
electron source/sink via neutrino absorptions and emissions by
free nucleons, respectively. The Lagrangian derivative is denoted
by d/dt 0/0t v = :. The tabulated realistic equation of state
based on relativistic mean eld theory (Shen et al. 1998) is imple-
mented according to the prescription in Kotake et al. (2003). The
mass accretion rate and the mass of the central object are xed to
be

M 1 M

s
1
and M
in
1.4 M

, respectively. The neutrino

heating is estimated under the assumptions that neutrinos are emit-
ted isotropically from the central object and that the neutrino ux
is not affected by local absorptions and emissions (see Ohnishi
et al. 2006). We consider only the interactions of electron-type
neutrinos and antineutrinos. Their temperatures are also assumed
to be constant and are set to be T
i
e
4 MeVand T
i
e
5 MeV,
typical values in the postbounce phase. The neutrino luminosity
is varied in the range of L
i
(6.0Y6.8) ; 10
52
ergs s
1
.
Spherical polar coordinates are adopted. In the radial direction,
the computational mesh is nonuniform, while the grid points are
equally spaced in other directions. We use 300 radial mesh points
to cover r
in
r r
out
, where r
in
$ 50 km is the radius of the in-
ner boundary, located roughly at the neutrino sphere, and r
out

2000 km is the radius of the outer boundary, at which the ow is
supersonic. A total of 30 polar and 60 azimuthal mesh points are
used to cover the whole solid angle. In order to see if this angular
resolution is sufcient, we have computed a model with the 300 ;
60 ; 120 mesh points and compared it to the counterpart with the
300 ; 30 ; 60 mesh points. As shown in Appendix B, the results
agree reasonably well with each other in both the linear and non-
linear phases. Although the computational cost does not allow us
to carry the convergence test further, we think that the resolution
of this study is good enough.
We use an articial viscosity of tensor type, described in Ap-
pendix A, instead of the von Neumann & Richtmyer type that
was originally employed in ZEUS-MP/2. For 3Dsimulations with
a spherical polar mesh, we nd the former preferable to prevent
the so-called carbuncle instability (Quirk 1994), which we ob-
serve around the shock front near the symmetry axis, 0 $ 0, .
With the original articial viscosity, an appropriate dissipation is
not obtained in the azimuthal direction for the shear ow result-
ing from the converging accretion, particularly when a ne mesh
is used (Stone & Norman 1992). We have also applied this ar-
ticial viscosity to axisymmetric 2Dsimulations and reproduced
the previous results (Ohnishi et al. 2006).
Figure 1 shows the radial distributions of various variables for
the unperturbed ows. The spherically symmetric steady accre-
tion ow through a standing shock wave is prepared in the same
manner as in Ohnishi et al. (2006). Behind the shock wave, the
electron fraction is less than 0.5 owing to electron capture, and a
region of negative entropy gradient with positive net heating
rates is formed for the neutrino luminosities L
i
(6.0Y6.8) ;
10
52
ergs s
1
. The values of these variables on the ghost mesh
points at the outer boundary are xed to be constant in time, while
on the ghost mesh points at the inner boundary they are set to be
identical to those on the adjacent active mesh points, except for the
radial velocity, which is xed to the initial value at both the inner
and outer boundaries.
In order to induce nonspherical instability, we have added a
radial velocity perturbation, cv
r
(r. 0. c), to the steady spherically
symmetric ow according to the equation
v
r
(r. 0. c) v
1D
r
(r) cv
r
(0. c). 6
where v
1D
r
(r) is the unperturbed radial velocity. In this study, we
consider three types of perturbations: (1) an axisymmetric (l 1,
m 0) single-mode perturbation,
cv
r
(0. c) /

3
4
_
cos 0 v
1D
r
(r). 7
(2) a nonaxisymmetric perturbation with l 1,
cv
r
(r. 0. c) /

3
4
_
cos 0

3
8
_
sin 0 cos c
_ _
v
1D
r
(r). 8
and (3) a random multi-mode perturbation,
cv
r
(0. c) / rand ; v
1D
r
(r) (0 rand < 1). 9
where rand is a pseudorandomnumber. The perturbation am-
plitude is set to be less than 1% of the unperturbed velocity. We
note that there is no distinction between m 1 and 1 modes
when the initial perturbation is added only to the radial velocity,
as is the case in this paper. To put it another way, the m 1
modes contribute equally. Hence, they are referred to as the
jmj 1 mode below. On the other hand, differences do showup,
IWAKAMI ET AL. 1208 Vol. 678
for example, if random perturbations are added to the nonradial
velocity components; this case is important in discussing the
origin of pulsar spins proposed by Blondin &Mezzacappa (2007),
and will be the subject of a forthcoming paper ( W. Iwakami
et al. 2008). We also note that the symmetry between m 1 and
1 modes is naturally removed if the unperturbed accretion ow
is rotating (see Yamasaki & Foglizzo 2008). All the models pre-
sented in this paper are summarized in Table 1.
In the next section, we perform the mode analysis as follows.
The deformation of the shock surface can be expanded as a linear
combination of the spherical harmonic components Y
m
l
(0. c):
R
S
(0. c)

1
l0

l
ml
c
m
l
Y
m
l
(0. c). 10
where Y
m
l
is expressed by the associated Legendre polynomial
P
m
l
and a constant K
m
l
, given as
Y
m
l
K
m
l
P
m
l
( cos 0)e
imc
. 11
K
m
l

2l 1
4
(l m)!
(l m)!

. 12
The expansion coefcients can be obtained as
c
m
l

_
2
0
dc
_

0
d0 sin 0 R
S
(0. c) Y
m
l
(0. c). 13
where the superscript asterisk (

) denotes complex conjugation.

3. RESULTS AND DISCUSSIONS
3.1. Axisymmetric Sinqle-Mode Perturbation
Before presenting the results of 3D simulations, we rst dem-
onstrate the validity of our newly developed 3D code. We com-
pare the axisymmetric ows obtained by 2D and 3D simulations
for the axisymmetric (l 1, m 0) single-mode perturbation.
By 2D simulations we mean that axisymmetry is assumed and
computations are done in the meridian section with all c deriva-
tives dropped, whereas in 3D simulations we retain all these de-
rivatives, and a 3Dmesh is employed. This validation is important
because numerical errors may induce azimuthal motions even
for the axisymmetric initial conditions. Hence, we have to con-
rmthat azimuthal errors do not appear in the nonaxisymmetric
simulation.
Figure 2 shows the time evolutions of the average shock ra-
dius R
S
for 2D ( Fig. 2a) and 3D ( Fig. 2b) results. The average
shock radius R
S
is obtained from the expansion coefcient c
0
0
in
TABLE 1
Summary of All the Models
Model Perturbation
Neutrino
Luminosity L
i
(10
52
ergs s
1
)
I ................... single-mode, l 1, m 0 6.0
II .................. multi-mode, l 1, m 0 and l 1, jmj 1 6.0
III................. single-mode, l 1, m 0 with
random perturbation at t 400 ms
6.0
IV ................ random perturbation 6.0
V.................. random perturbation 6.4
VI ................ random perturbation 6.8
VII ............... axisymmetric random perturbation 6.0
VIII.............. axisymmetric random perturbation 6.8
IX ................ none 6.0
X.................. none 6.8
Fig. 1.Proles of various variables for the unperturbed spherically symmetric accretion ows with different neutrino luminosities.
3D SIMULATIONS OF SASI IN SUPERNOVAE 1209 No. 2, 2008
equation (10) multiplied by K
0
0
. The solid, dashed, and dotted lines
correspond to the neutrino luminosities L
i
of 6.0, 6.4, and 6.8 ;
10
52
ergs s
1
, respectively. One cannot expect a perfect agreement
between two computations of the exponential growth of the insta-
bility followed by turbulent motions through mode coupling, as
considered here; however, it is still obvious that the results of the
2Dand 3Dsimulations agree in essential features. In particular, in
both the 2D and 3D results, we see that the R
S
is settled to an al-
most constant value for L
i
6.0 and 6.4 ; 10
52
ergs s
1
, whereas
it continues to increase for L
i
6.8 ; 10
52
ergs s
1
.
Figure 3 presents the normalized amplitudes jc
m
l
/c
0
0
j as a func-
tion of time for model I (L
i
6.0 ; 10
52
ergs s
1
) for 2D(Fig. 3a)
and 3D (Fig. 3b) results. The red, blue, black, and gray solid lines
correspond to the modes of (l. m) (1. 0), (2,0), (3,0), and (4,0),
respectively. As can be seen, the time evolution can be divided into
two phases. First is the linear growth phase, in which the amplitude
of the initially imposed mode with (l. m) (1. 0) grows exponen-
tially. This lasts for $100 ms. Higher modes are also generated by
mode couplings and grow exponentially during this phase. Then
starts the second phase, which is characterized by the saturation of
amplitudes of the order of 0.1. In this phase, the accretion ow be-
comes turbulent. It is interesting to note that in this nonlinear satura-
tion phase the amplitude of the mode with (l. m) (2. 0) is almost
of the same order as the initially imposed mode with (l. m) (1. 0)
Fig. 2.Time evolutions of the average shock radius R
S
in 2D (left) and 3D (riqht) for the axisymmetric l 1, m 0 single-mode perturbation.
Fig. 3.Time evolutions of the normalized amplitudes jc
m
l
/c
0
0
j for model I with the axisymmetric l 1, m 0 single-mode perturbation in 2D (left) and 3D (riqht).
[See the electronic edition of the Journal for a color version of this qure.]
IWAKAMI ET AL. 1210 Vol. 678
and is dominant over other modes, which fact was observed in
Ohnishi et al. (2006).
Most important of all, however, is the fact that none of the
m 6 0 modes are produced, implying that the perturbed ow re-
tains axisymmetry, which is a necessary condition for the numer-
ical study of nonaxisymmetric instability. Although the results
of the 2D and 3D simulations are not identical, the essential fea-
tures such as the linear growth rate of the initial perturbation
(l 1, m 0) and the production of other modes via nonlinear
mode coupling, as well as the saturation levels in the nonlinear
phase, are in good agreement for the two cases. The quantita-
tive differences between the 2Dand 3Dresults mainly originate
from the difference in the time steps, which depend on the mini-
mumgrid width. Note that in additionto the 300radial and30 polar
mesh points used for the 2D computations, 60 azimuthal mesh
points are employed in the 3D simulations and, as a result, smaller
time steps are usually taken for the 3D computations.
3.2. Nonaxisymmetric Perturbation with l 1
Now we discuss the results of genuinely 3D simulations, in
which the nonaxisymmetric perturbation with (l 1, jmj 1) is
added to the axisymmetric (l 1, m 0) perturbation.
Figure 4 shows the time evolutions of some of the normalized
amplitudes jc
m
l
/c
0
0
j for model II with the neutrino luminosity L
i

6.0 ; 10
52
ergs s
1
. The red, yellow, blue, light blue, green, black,
brown, violet, and pink solid lines denote the amplitudes of the
modes with (l. jmj) (1. 0), (1,1), (2,0), (2,1), (2,2), (3,0), (3,1),
(3,2), and (3,3), respectively. In the linear phase, the initially im-
posed modes with (l. jmj) (1. 0) and (1,1) grow exponentially,
just as in the axisymmetric single-mode perturbation. It should be
noted that the growth rate of the (l. jmj) (1. 1) mode is identical
to that of the (l. m) (1. 0) mode. This is just as expected for the
spherically symmetric background, for which modes with the
same l but different mare degenerate in the linear eigenfrequency.
It can also be seen fromFigure 4 that the modes with (l. jmj)
(2. 0), (2,1), (2,2) are rst produced by the nonlinear mode cou-
plings of (l. jmj) (1. 0), (1,1). Then the modes with (l. jmj)
(3. 0), (3,1), (3,2), (3,3) can be produced by the couplings of the
rst-order modes with (l. jmj) (1. 0), (1,1) and the second-
order modes with (l. jmj) (2. 0), (2,1), (2,2). Even higher order
modes are produced subsequently toward the nonlinear satura-
tion, but they are not shown here. Although the branching ratios
should be investigated in more detail before coming to any con-
clusions, the above sequence of mode generation strongly sug-
gests that the nonlinear coupling is mainly of quadratic nature.
In the nonlinear phase that begins at t $ 200 ms, these mode
amplitudes are saturated and settled into a quasi-steady state. As
in the axisymmetric case, the l 1 modes both with m 0 and
jmj 1 are dominant over other modes in this stage for the non-
axisymmetric case. One thing to be noted here, however, is the
fact that the saturated amplitudes for model II are lower than
those for model I in general (see Fig. 3b). This is, we think, be-
cause the turbulent energy is shared by a larger number of modes
in the nonaxisymmetric case than in the axisymmetric case. To
demonstrate this more clearly, we have added a random pertur-
bation to the radial velocity in the axisymmetric model I at t
400 ms, by which time the axisymmetric nonlinear turbulence
has fully grown. We refer to this model as model III. The time
evolutions of the normalized amplitudes jc
m
l
/c
0
0
j for model III are
shown in Figure 5. It is clear that the axisymmetric m 0 mode
amplitudes are reduced after the additional perturbation is given,
and the nonaxisymmetric m 6 0 modes grow to saturation level
at t $ 450 ms. The power spectra of the turbulent motions will
be discussed in more detail later.
3.3. Random Multi-Mode Perturbations
3.3.1. Dynamical Features and Critical Luminosity
Having understood the elementary processes of the linear
growth and nonlinear mode couplings in the previous section,
we now discuss the 3D SASI induced by random multi-mode
perturbations, which are supposed to be closer to reality. In this
subsection, we showthe typical dynamics, paying attention to the
time evolution of the shock wave, and hence to the critical lumi-
nosity, at which the stalled shock is revived.
Figure 6 shows the time variations of the average shock radius
R
S
for models IV, V, and VI, with neutrino luminosities L
i

6.0, 6.4, and 6.8 ; 10
52
ergs s
1
, respectively. It is found that for
L
i
6.0 and 6.4 ; 10
52
ergs s
1
, the shock settles to a quasi-
steady state after the linear growth, whereas it continues to expand
for L
i
6.8 ; 10
52
ergs s
1
, which is also true for the axisym-
metric counterpart displayed in the right panel of Figure 6 for
comparison. This implies that the critical neutrino luminosity is
not much affected by the change from 2D to 3D. In the follow-
ing discussion we refer to the models with L
i
6.0 and 6.4 ;
10
52
ergs s
1
as the non-explosion models, and the model with
Fig. 4.Time variations of the normalized amplitudes jc
m
l
/c
0
0
j for model II, in
which the modes with (l. jmj) (1. 0). (1. 1) are initially imposed. [See the elec-
tronic edition of the Journal for a color version of this qure.]
Fig. 5.Time evolutions of the normalized amplitudes jc
m
l
/c
0
0
j for model III, in
which we have added a random perturbation to the radial velocity at t 400 ms.
[See the electronic edition of the Journal for a color version of this qure.]
3D SIMULATIONS OF SASI IN SUPERNOVAE 1211 No. 2, 2008
L
i
6.8 ; 10
52
ergs s
1
as the explosion model, and we look
into their dynamical features in turn. The mode analyses will be
done in xx 3.3.2 and 3.3.3.
Figure 7 shows the side views of the iso-entropy surfaces to-
gether with the velocity vectors in the meridian section at four
different times for the non-explosion model with L
i
6.0 ;
10
52
ergs s
1
(model IV). The hemispheres (0 c ) of eight
iso-entropy surfaces are superimposed on one another, and the
outermost surface almost corresponds to the shock front. The ini-
tial perturbations grow exponentially in the linear phase (Fig. 7a;
t 40 ms). At the end of the linear phase, a blob of high entropy
is formed, which subsequently grows (Fig. 7b; t 93 ms). High-
entropy blobs are continuously generated, and the nonlinear phase
begins (Fig. 7c; t 100 ms). Circulating ows occur inside the
blobs, while high-velocity accretions onto a protoYneutron star
surround the blobs. These blobs expand and shrink repeatedly,
being distorted, split, and merged with each other inside the shock
wave (Fig. 7d; t 350 ms). Reecting these complex motions,
the shock surface oscillates in all directions but remains almost
spherical for the nonaxisymmetric model. This is in sharp contrast
to the axisymmetric case, in which large-amplitude oscillations
occur mainly in the direction of the symmetry axis (Blondin et al.
2003; Ohnishi et al. 2006).
Figure 8 displays snapshots of the iso-entropy surfaces and
the velocity vectors for the explosion model with L
i
6.8 ;
10
52
ergs s
1
(model VI ). As in the non-explosion model, the se-
quence of events starts with the linear growth of the initial pertur-
bation inside the shock wave (Fig. 8a; t 40 ms). In the explosion
model, however, many high-entropy blobs emerge simultaneously
much earlier on ( Fig. 8b; t 70 ms) than in the non-explosion
model. These blobs then repeatedly expand and contract, being
distorted, split, and merged together just as in the non-explosion
counterpart ( Fig. 8c; t 80 ms). After that, some of the blobs
get bigger, absorbing other blobs ( Fig. 8d; t 350 ms) and, as a
result, the shock radius increases almost continuously up to the
end of the computation (t 400 ms), as already demonstrated in
Figure 6a. At this point, the shock surface looks ellipsoidal rather
than spherical. However, the major axis is not necessarily aligned
with the symmetry axis, nor is the ow is symmetric with respect
to this major axis.
3.3.2. Mode Spectra
Next we look into the spectral intensity in more detail. As a
standard case, we take model IV with L
i
6.0 ; 10
52
ergs s
1
.
Figure 9 shows the time evolutions of the normalized amplitudes
jc
m
l
/c
0
0
j with m 0 and compares them with the axisymmetric
counterparts in model VII. Note that the m 6 0 modes also exist
in the nonaxisymmetric model; these are not shown in the gure
but will be discussed in the following paragraphs. As can be
clearly seen, the amplitude of each mode grows exponentially
until $100 ms, which is the linear phase. Note in particular that
the growth rate and oscillation frequency for the l 1 mode are
the same as those obtained for the model with the single-mode
perturbation. After $100 ms, the mode amplitudes are saturated
and the evolution enters the nonlinear phase, with a clear domi-
nance of the modes with l 1. 2. It is also evident in the gure that
the saturation level is lower in general for the nonaxisymmetric case
than for the axisymmetric one, which conrms the claimin the pre-
vious section based on the results for the single-mode perturbation.
In Figure 10, we display snapshots at four different times
of the spectral distributions for both the nonaxisymmetric (left
panels) and axisymmetric (riqht panels) cases. The top panels
correspond to the linear phase, in which the intensity is distributed
rather uniformly over all modes. As time passes, however, the am-
plitudes of low-l modes grow much more rapidly than those with
higher l (second and third rows). One can see a similarity in the
time evolutions between the nonaxisymmetric and axisymmetric
models. It should be noted again that there is no superiority in the
linear growth rate among different mmodes in the nonaxisymmet-
ric case. Since the amplitudes of different modes are oscillating in
time with some phase lags, the mode with (l. jmj) (1. 0) is larg-
est at one instance (Fig. 10b; t 30 ms), whereas the mode with
(l. jmj) (1. 1) has the greatest amplitude at another instance
( Fig. 10b; t 60 ms). On average, however, none of them is
superior to others. This is also true of the explosion models
described in the next section. After the nonlinear phase starts
Fig. 6.Time evolutions of the average shock radius R
S
for models IV, Vand VI, in which a random multi-mode perturbation is imposed in 3D (left) and 2D (riqht).
IWAKAMI ET AL. 1212 Vol. 678
(bottompanels), the growths of all modes are saturated, and the
spectral distributions are settled to be quasi-steady. It is again
obvious in these panels that the low-l modes are dominant in
the nonlinear phase, and the saturation level is lower in the non-
axisymmetric case.
Now we turn our attention to the quasi-steady turbulence in
the nonlinear phase. Shown as a function of l in Figure 11 are the
power spectra jc
m
l
/c
0
0
j
2
averaged over the interval from t 150
to 400 ms. The nonaxisymmetric and axisymmetric cases are
shown by open and lled circles, respectively. In the left panel,
modes with different m are plotted separately, whereas they are
summed in the right panel. We nd that the time-averaged power
spectra are not very different among the modes with the same
l but different m. This implies that the equipartition of the tur-
bulence energy is nearly established among modes with the same
l. This will have important ramications for the origin of pulsar
spin, and will be discussed in our forthcoming paper (W. Iwakami
et al. 2008, in preparation).
The right panel of Figure 11 demonstrates that the time-averaged
power spectrum summed over m for the nonaxisymmetric case is
Fig. 7.Iso-entropy surfaces and velocity vectors in the meridian section for model IV. The hemispheres (0 c ) of eight iso-entropy surfaces are superimposed,
with the outermost surface nearly corresponding to the shock front. [See the electronic edition of the Journal for a color version of this qure.]
3D SIMULATIONS OF SASI IN SUPERNOVAE 1213 No. 2, 2008
not very different fromthat for the axisymmetric counterpart. This
means that the total turbulence energy does not differ very much
between the axisymmetric and nonaxisymmetric cases. As a re-
sult, the power of each mode in the nonaxisymmetric case is
smaller by roughly a factor of 2l 1 than that of the same l mode
in the axisymmetric case. This is the reason why we have ob-
served that the saturation level of the nonlinear SASI is smaller
in the nonaxisymmetric case than in the axisymmetric case and
the shock surface oscillates in all directions with smaller ampli-
tudes in the nonaxisymmetric ow, whereas it sloshes in the di-
rection of the symmetry axis with larger amplitudes in the axi-
symmetric case. The difference in the uctuations of the average
shock radius seen in Figures 6a and 6b can also be explained in
the same manner.
Another interesting feature observed in Figure 11 is the fact
that the time-averaged power spectra obey a power law at l k10
for both the nonaxisymmetric and axisymmetric models. Two
straight lines in Figure 11a are the ts to the data for l ! 10, each
obtained for the axisymmetric and nonaxisymmetric models. The
powers are found to be 5.7 and 4.3 for the nonaxisymmetric
Fig. 8.Iso-entropy surfaces and velocity vectors in the meridian section for the explosion model (model VI ). Note that the displayed region is 2.5 times larger for
panel d. [See the electronic edition of the Journal for a color version of this qure.]
IWAKAMI ET AL. 1214 Vol. 678
and axisymmetric cases, respectively. Although the difference is
almost unity, which originates from the difference in multiplicity
of the modes with the same l, the slope excluding this effect is still
a little bit steeper in the nonaxisymmetric case, as seen in Fig-
ure 11b. At the moment we do not knowif this difference in slope
is real or not, and the power itself also remains to be explained
theoretically.
3.3.3. The Explosion Models
So far, we have been discussing the non-explosion models, in
which the SASI is saturated in the nonlinear phase and is settled
to a quasi-steady state. In considering the possible consequences
of SASI after a supernova explosion occurs, however, it is also
interesting to investigate the explosion models, in which a shock
revival appears as a result of SASI.
Figure 12 shows the time evolutions of the normalized ampli-
tudes jc
m
l
/c
0
0
j for the explosion model (model VI ) and compares
them with the axisymmetric counterpart (model VIII ). One fea-
ture shared by both the axisymmetric and nonaxisymmetric explo-
sion models is that the nonlinear stage is divided into two phases.
The earlier phase is quite similar to the nonlinear phase that we
have seen so far for the non-explosion models. The later phase, on
the other hand, has the distinctive feature that the l 1 mode be-
comes much more remarkable than other modes, and the oscil-
lation period tends to be longer as the shock radius gets larger and
the advection-acoustic cycle, supposedly an underlying mecha-
nism of the instability, takes longer. The later prominence of the
l 1 mode is intriguing, and will be important to any quantitative
discussion of the pulsar kick velocity. Atheoretical account, how-
ever, is yet to be given. The differences between the nonaxisym-
metric and axisymmetric models, such as the saturation level, are
carried over to the explosion models.
The power spectra averaged over time intervals of 100 t
200 ms and 250 t 400 ms are presented in Figure 13a and
13b, respectively. As in the non-explosion models, the equiparti-
tion among different m modes is again established for the explo-
sion models. This is true even in the later nonlinear phase, as seen
in the right panel of the gure. As a result, the saturation level is
smaller for the nonaxisymmetric case than for the axisymmetric
case, as mentioned above. The power spectra in the earlier non-
linear phase look very similar to those found in the non-explosion
models, with the power law being satised for l k10. In the later
nonlinear phase, on the other hand, the power law is extended to
much lower l. This is related to the late-time prominence of the
l 1 mode, as mentioned above. In fact, the amplitudes in the
lower l portion of the power spectrumare enhanced in 250 t
400. The mechanism of this amplication remains to be under-
stood, but it might be related to the volume-lling thermal con-
vection advocated for the late stage of convective instability in a
supernova core by Kifonidis et al. (2006). As a result of this evo-
lution of the spectrum, the shock tends to be ellipsoidal as the
shock expands.
3.4. Neutrino Heatinq
The SASI is supposed to be an important ingredient not only
for a pulsars proper motion and spin, but also for the explosion
mechanism itself. In this section, we look into neutrino heating
in nonaxisymmetric SASI.
In Figure 14 we showthe color contours of the net heating rate
in the meridian section for the non-explosion model with L
i

6.0 ; 10
52
ergs s
1
(nonaxisymmetric model IVand axisymmet-
ric model VII ), as well as for the nonaxisymmetric explosion
model with L
i
6.8 ; 10
52
ergs s
1
(model VI ). In the early
phase, a cooling region with negative net heating rates exists
around the protoYneutron star, while the heating region extends
over the cooling region up to the stalled shock wave in all cases.
As time passes and the SASI grows, a pocket of regions with
positive but relatively low net heating rates appear. These regions
correspond to the high-entropy blobs (high-entropy rings for the
axisymmetric case), where the circulating owexists, as observed
in Figures 7b, 7c, and 7d. Since the neutrino emission in these re-
gions is more efcient than in the surroundings, the net heating
rate is rather low.
Althoughthe critical neutrinoluminosities are not very different
between the nonaxisymmetric and axisymmetric cases, the spatial
distributions of neutrino heating are different. In the axisymmetric
Fig. 9.Time evolutions of the normalized amplitudes jc
m
l
/c
0
0
j for models IV(left) and VII (riqht). Note that the m 6 0 modes also exist in the nonaxisymmetric model,
but are not shown in this gure. [See the electronic edition of the Journal for a color version of this qure.]
3D SIMULATIONS OF SASI IN SUPERNOVAE 1215 No. 2, 2008
Fig. 10.Normalized amplitudes jc
m
l
/c
0
0
j for model IV (left panels) and model VII (riqht panels) at different times. Note that the time-averaged values are plotted in
panels d and h. [See the electronic edition of the Journal for a color version of this qure.]
case, the shock wave oscillates up and down, whereas it moves in
all directions in the nonaxisymmetric case. The oscillation am-
plitudes are larger in the axisymmetric than in the nonaxisym-
metric case in general, as repeatedly mentioned. Reecting this
difference in the shock motions, the neutrino heating is enhanced
mainly in the polar regions in the axisymmetric case, while in the
nonaxisymmetric case the heating rate is affected by SASI chiey
around the high-entropy blobs. These effects are clearly seen in
Figure 14 (see also Fig. 7).
As described in x 3.3.1, the generation of the high-entropy
blobs starts around the end of the linear phase and continues dur-
ing the nonlinear phase. Although the blobs repeatedly expand
and contract, sometimes merging or splitting, the turbulent mo-
tions together with the neutrino heating nally settle to a quasi-
steady state in the non-explosion model. For the explosion model,
on the other hand, the number and volume of high-entropy blobs
increase much more quickly, and as a result, the heating region
prevails, pushing the shock wave outward and narrowing the cool-
ing region near the protoYneutron star.
Figure 15 shows the time evolutions of the net heating rates
integrated over the gain region inside the shock wave. For the non-
explosion model (model IV; solid line), the volume-integrated
heating rate grows until t $ 150 ms, but is saturated there-
after, whereas it increases continuously for the explosion model
Fig. 11.Time-averaged power spectra for models IV and VII. The average is taken over t 150Y400 ms. In the left panel, modes with different m are plotted
separately; they are summed up in the right panel.
Fig. 12.Time evolutions of the normalized amplitudes jc
m
l
/c
0
0
j for the explosion models model VI (left) and model VIII (riqht). [See the electronic edition of the
Journal for a color version of this qure.]
3D SIMULATIONS OF SASI IN SUPERNOVAE 1217
(model VI; dash-dotted line). These behaviors of the volume-
integrated heating rate are in accord with the time evolutions of the
average shock radius, as shown in Figure 6a. It is clear from the
gure that the heating rates are mainly affected by SASI during the
nonlinear phase. For comparison, the spherically symmetric coun-
terparts, models IX with L
i
6.0 ; 10
52
ergs s
1
and X with
L
i
6.8 ; 10
52
ergs s
1
, are also presented as dashed and dotted
lines, respectively. Note that the last model does not lead to shock
revival even with this high neutrino luminosity. It is found that the
SASI enhances the neutrino heating for both the explosion and
non-explosion models.
It is understandably a difcult task to make a clean comparison
with more realistic simulations, since they are highly dynamical,
and the extraction of an unperturbed background is not easy. The
SASI they observe is more often than not nonlinear from the be-
ginning. We do, however, point out that the heating rate of our
model IVis similar to what was observed in the nonrotating model
of Marek & Janka (2007; see the line for HW-2D in the middle
panel of Fig. 7) after t $ 200 ms. Hence, our results with simpli-
ed physics are consistent with their results taking more detailed
physics into account, at least for the nonlinear phase. The linear
phase is more difcult to compare, since their models probably
have no linear phase at all. This clearly demonstrates the com-
plimentary roles of the idealized toy models and detailed
simulations.
4. CONCLUSIONS
In this paper we have studied the nonaxisymmetric SASI by
3D hydrodynamical simulations, taking into account a realistic
EOS and neutrino heating and cooling. We have added various
nonaxisymmetric perturbations to spherically symmetric steady
ows that accrete through a standing shock wave onto a protoY
neutron star, being irradiated by neutrinos emitted fromthe protoY
neutron star. Mode analysis has been done for the deformation of
the shock surface by the spherical harmonics expansion. After
conrming that our 3D code is able to reproduce for the axisym-
metric perturbations the previous results for 2D SASI that we
obtained in Ohnishi et al. (2006), we have done genuinely 3D
simulations and found the followings.
First, the model of the initial perturbation with (l. jmj) (1. 0)
and (1,1) has demonstrated that the nonaxisymmetric SASI pro-
ceeds much in the same manner as the axisymmetric SASI: the
linear phase, in which the initial perturbation grows exponen-
tially, lasts for about 100 ms and is followed by a nonlinear phase,
in which various modes are produced by nonlinear mode cou-
pling, and their amplitudes are saturated. It has been found that
the critical neutrino luminosity, above which shock revival occurs,
is not very different between 2Dand 3D. For neutrino luminosities
lower than the critical value, the SASI settles to a quasi-steady tur-
bulence. We have found that the saturation level of each mode in
the nonaxisymmetric SASI is lower in general than that of its axi-
symmetric counterpart. This is mainly due to the fact that the num-
ber of the modes contributing to the turbulence is larger for the
nonaxisymmetric case. The sequence of mode generation, on the
other hand, strongly suggests that nonlinear mode coupling is
chiey quadratic in nature.
Second, the simulations with random multi-mode perturba-
tions imposed initially have shown that the dynamics in the lin-
ear phase is essentially a superposition of those of single modes.
Toward the end of the linear phase, high-entropy blobs are gen-
erated continuously and grow, starting the nonlinear phase. We
have observed that these blobs repeatedly expand and contract,
merging and splitting from time to time. In the non-explosion
models, these nonlinear dynamics lead to the saturation of mode
amplitudes and quasi-steady turbulence. For the explosion mod-
els, on the other hand, the production of blobs proceeds much
more quickly, followed by an oligarchic evolution, with a rela-
tively small number of large blobs absorbing smaller ones, and
as a result the shock radius increases almost monotonically. The
spectral analysis has clearly demonstrated that low-l modes are
predominant in the nonlinear phase, just as in the axisymmetric
case. We have also shown that equipartition is nearly established
among different mmodes in the quasi-steady turbulence, and that
the spectrum summed over m in the nonaxisymmetric case is
quite similar to the axisymmetric counterpart. This implies that
the larger number of modes in the nonaxisymmetric case is the
main reason why the amplitude of each mode is smaller in 3D
than in 2D. The power spectrumis approximated by a power law
Fig. 13.Time-averaged power spectra for the explosion models, model VI and VIII. The averages are taken for (a) t 100Y200 ms and (b) t 250Y400 ms.
IWAKAMI ET AL. 1218
Fig. 14.Color contours of the net heating rate in the meridian section for model IV(top panels), model VII (middle panels), and model VI (bottompanels) at different
times. All panels have the same spatial scale. [See the electronic edition of the Journal for a color version of this qure.]
for l k10. Although the slope is slightly steeper for the nonaxi-
symmetric models, whether the difference is signicant or not is
unknown at present.
We have seen in the explosion models, on the other hand, that
the oscillation period of each mode becomes longer in the late
nonlinear phase, as the shock radius gets larger and the advec-
tion-acoustic cycle becomes longer. What is more interesting is
the fact that in this late phase the dominance of low l modes
becomes even more remarkable. Although this may be related to
the volume-lling thermal convection, the mechanism has yet to
be revealed. This spectral evolution leads to the global deforma-
tion of the shock surface to an ellipsoidal shape, whose major
axis is not, however, necessarily aligned with the symmetry axis.
Finally, we have presented the neutrino heating in 3DSASI. It
has been shown that the volume-integrated heating rate is affected
mainly in the nonlinear phase. Acomparisonwith spherically sym-
metric counterparts has conrmed that the SASI also enhances
neutrino heating in the nonaxisymmetric case. Although the crit-
ical neutrino luminosity in the nonaxisymmetric SASI is not much
changed from that for the axisymmetric case, the spatial distri-
bution of neutrino heating is different in the nonlinear phase.
Relatively narrow regions surrounding high-entropy blobs are
efciently heated for the nonaxisymmetric case, while wider re-
gions near the symmetry axis are heated strongly, in accord with
the sloshing motion of shock wave along the symmetry axis, for
the axisymmetric case. For the non-explosion models, the high-
entropy blobs produced by neutrino heating occupy the inside
of shock wave, repeatedly expanding and contracting and being
intermittently split and merged, but the ow nally settles to a
quasi-static state. For the explosion models, on the other hand, the
high-entropy blobs are generated much more quickly and extend
the heating region, pushing the shock outward and narrowing the
cooling region near the protoYneutron star.
In this paper we have not discussed the instability mechanism,
which is still controversial at the moment. Based on our previous
work (Ohnishi et al. 2006), we prefer the advection-acoustic cycle
proposed by Foglizzo (2001, 2002) to the purely acoustic cycle
discussed by Blondin &Mezzacappa (2006). It is, however, fair to
mention that most of our models have ratios of the shock radius to
the inner boundary radius for which the recent analysis by Laming
(2007) predicts the operation of a pure acoustic cycle. Since his
analysis includes some approximations, it is certainly not the nal
verdict in our opinion, and further investigations are needed.
In the present study we have not observed a persistent segre-
gation of angular momentumin the accretion ow, such as found
by Blondin & Mezzacappa (2007) and Blondin & Shaw (2007),
although instantaneous spiral ows are frequently seen. As dis-
cussed above, equipartition is nearly established among different
mmodes in our models. It should be emphasized here again, how-
ever, that we have added the initial perturbations only to the radial
velocity in this study, and as a result, modes with m 1 are
equally contributing. We defer the analysis of models with initial
nonaxisymmetric perturbations added also to the azimuthal com-
ponent of the velocity to a forthcoming paper ( W. Iwakami et al.
2008, in preparation), in which we will also discuss a possible
correlation between the kick velocity and spin of neutron stars if
they are indeed produced by the 3D SASI. Many questions re-
garding the SASI still remain to be studied; in particular, the inu-
ences of rotation and magnetic eld are among the top priorities,
and will be addressed elsewhere in the near future.
W. I. expresses her sincere gratitude to Kazuyuki Ueno,
Michiko Furudate, and the members of the Sawada and Ueno
laboratory for continuing encouragement and advice. She would
also like to thank Yoshitaka Nakashima of Tohoku University
for his advice on visualization by AVS. K. K. expresses his thanks
to Katsuhiko Sato for his continuous encouragement. Numerical
computations were performed on the Altix3700Bx2 at the Insti-
tute of Fluid Science, Tohoku University, and on VPP5000 and
the general common use computer system at the Center for Com-
putational Astrophysics (CfCA), National Astronomical Observa-
tory of Japan. This study was partially supported by the Program
for Improvement of Research Environment for Young Researchers
fromSpecial Coordination Funds for Promoting Science and Tech-
nology (SCF), Grants-in-Aid for Scientic Research (S19104006,
S14102004, 14079202, 14740166), and a Grant-in-Aid for the 21st
century COE program Holistic Research and Education Center
for Physics of Self-organizing Systems of Waseda University
by the Ministry of Education, Culture, Sports, Science, and Tech-
nology (MEXT) of Japan.
APPENDIX A
TENSOR ARTIFICIAL VISCOSITY
Here we present the tensor-type articial viscosity, Q, that we use in this paper. It is a direct extension to 3D of that employed by
Stone & Norman (1992) for 2D simulations. Following the notations of Stone & Norman (1992), Q can be written as
Q
l
2
,:= v :v
1
3
: = v e
_ _
if := v < 0.
0 otherwise.
_
_
_
A1
Fig. 15.Time variations of the net heating rate integrated over the gain re-
gion inside the shock wave. Models IVand VI are compared with the spherically
symmetric counterparts, models IX and X.
IWAKAMI ET AL. 1220 Vol. 678
where l denotes a constant with a dimension of length, :v (v
i. j
v
j. i
)/2 is the symmetrized velocity-gradient tensor, and e is the unit
tensor. Dropping the off-diagonal components, we utilize only the diagonal components of Q in this study, which are written in nite
difference form as
Q11
i. j. k
l
2
i. j. k
d
i. j. k
:= v
i. j. k
:v
(11)
_ _
i. j. k

1
3
:= v
i .j. k
_ _
.
Q22
i. j. k
l
2
i. j. k
d
i. j. k
:= v
i. j. k
:v
(22)
_ _
i. j. k

1
3
:= v
i. j. k
_ _
.
Q33
i. j. k
l
2
i. j. k
d
i. j. k
:= v
i. j. k
:v
(33)
_ _
i. j. k

1
3
:= v
i. j. k
_ _
. A2
where d
i. j. k
stands for the density at the site specied by (i. j. k), and l
i. j. k
is written as
l
i. j. k
max (C
2
dx1a
i
. C
2
dx2a
j
. C
2
dx3a
k
). A3
Here C
2
is a dimensionless constant controlling the number of grid points, over which a shock is spread, and dx1a
i
, dx2a
j
, and dx3a
k
are
the grid widths of the a-mesh at the i, j, and kth grid points in the r, 0, and c directions, respectively. The a-mesh and b-mesh are dis-
tinguished in ZEUS-MP/2 and are dened on the cell edge and cell center, respectively (see Stone & Norman 1992 for more details).
The articial viscous stress := Q and articial dissipation Q : :v in the momentum equation (2) and energy equation (3) are
given as
:= Q
(1)

1
q
2
2
q
31
0
0x
1
q
2
2
q
31
Q
11
_ _
.
:= Q
(2)

1
q
2
q
2
32
0
0x
2
q
2
32
Q
22
_ _

Q
11
q
2
q
32
0q
32
0x
2
.
:= Q
(3)

1
q
2
q
32
0Q
33
0x
3
. A4
Q : :v l
2
,:= v
1
3
:v
(11)
:v
(22)
_ _
2
:v
(11)
:v
(33)
_ _
2
:v
(33)
:v
(22)
_ _
2
_ _
. A5
where q
2
r, q
31
r, and q
32
sin 0 are the metric components for the spherical coordinates, (x
1
. x
2
. x
3
) (r. 0. c) (see again Stone
& Norman 1992). These equations are discretized as
:= Q
1. i. j. k

q2b
2
i
q31b
i
Q11
i. j. k
q2b
2
i1
q31b
i1
Q11
i1. j. k
q2a
2
i
q31a
i
dx1b
i
.
:= Q
2.i. j. k

q32b
2
j
Q22
i. j. k
q32b
2
j1
Q22
i. j1. k
q2b
i
q32a
2
i
dx2b
j

Q11
i. j. k
Q11
i. j1. k
2q2b
i
q32a
j
0q32a
j
0x
2
_ _
.
:= Q
3.i. j. k

Q33
i. j. k
Q33
i. j. k1
q2b
i
. q32b
j
dx3b
k
. A6
(Q : :v)
i. j. k
l
2
i. j. k
d
i. j. k
( := v)
i. j. k
1
3
_
( :v
(11)
)
i. j. k
( :v
(22)
)
i. j. k
_ _
2
( :v
(11)
)
i. j. k
( :v
(33)
)
i. j. k
_ _
2
( :v
(33)
)
i. j. k
( :v
(22)
)
i. j. k
_ _
2
_
. A7
where q2a, q31a, and q32a are q
2
, q
31
, and q
32
dened on the a-mesh, and q2b, q31b, and q32b are those dened on the b-mesh. The
terms dx1b
i
, dx2b
j
, and dx3b
k
represent the width of the b mesh at the i, j, and kth grid points in the r, 0, and c directions, respectively.
Finally, the velocity-gradient tensor ( :v
(11)
)
i. j. k
. ( :v
(22)
)
i. j. k
, and ( :v
(33)
)
i. j. k
are given by
:v
(11)
_ _
i. j. k

v1
i1. j. k
v1
i. j. k
dx1a
i
.
:v
(22)
_ _
i. j. k

v2
i. j1. k
v2
i. j. k
q2b
i
dx2a
j

v1
i. j. k
v1
i1. j. k
2q2b
i
0q2b
i
0x
1
_ _
.
:v
(33)
_ _
i. j. k

v3
i. j. k1
v3
i. j. k
q2b
i
q32b
j
dx3a
k

v2
i. j. k
v2
i. j1. k
2q31b
i
q32b
j
0q32b
j
0x
2
_ _

v1
i. j. k
v1
i1. j. k
2q31b
i
0q31b
i
0x
1
_ _
. A8
3D SIMULATIONS OF SASI IN SUPERNOVAE 1221 No. 2, 2008
APPENDIX B
NUMERICAL CONVERGENCE TESTS
In order to see if the numerical resolution employed in the main body is sufcient, we increase the number of angular grid points and
compare the results. Figure 16a shows the time evolutions of the normalized amplitudes jc
m
l
/c
0
0
j in the linear phase. In this comparison,
we impose the l 1, jmj 1 perturbation initially. We refer to the model with 300 ; 30 ; 60 mess points as MESH0, and to that with
300 ; 60 ; 120 grid points as MESH1 in the gure. We nd that the linear growth rates agree with each other reasonably well, al-
though the coarser mesh slightly overestimates the growth time. Figure 16b presents the power spectra jc
m
l
/c
0
0
j
2
that are time-averaged
over the nonlinear phase. The random perturbation is imposed in this case. It is again clear that the results for MESH0 are in good
agreement with those for MESH1.
It should be mentioned that for MESH0 it takes 32 parallel processors about 1.5 days to compute the evolution up to t 400 ms,
while MESH1 needs 22 days even for 128 parallel processors. This is partly because of the difference in the Courant numbers, which
are set to 0.5 for MESH0 but 0.1 for MESH1 to better use the tensor-type articial viscosity. Although this severe limit of CPU time
does not allowus to do more thorough convergence tests, we think, based on the results of the tests shown above, that our results given
in this paper are credible.
REFERENCES
Blondin, J. M., & Mezzacappa, A. 2006, ApJ, 642, 401
. 2007, Nature, 445, 58
Blondin, J. M., Mezzacappa, A., & DeMarino, C. 2003, ApJ, 584, 971
Blondin, J. M., & Shaw, S. 2007, ApJ, 656, 366
Burrows, A., Hayes, J., & Fryxell, B. A. 1995, ApJ, 450, 830
Burrows, A., Livne, E., Dessart, L., Ott, C. D., & Murphy, J. 2006, ApJ, 640,
878
Foglizzo, T. 2001, A&A, 368, 311
. 2002, A&A, 392, 353
Foglizzo, T., Scheck, L., & Janka, H.-T. 2006, ApJ, 652, 1436
Hayes, J. C., Norman, M. L., Fiedler, R. A., Bordner, J. O., Li, P. S., Clark, S.
E., ud-Doula, A., & Mac Low, M.-M. 2006, ApJS, 165, 188
Herant, M., Benz, W., Hix, W. R., Fryer, C. L., & Colgate, S. A. 1994, ApJ,
435, 339
Janka, H.-T., & Mueller, E. 1996, A&A, 306, 167
Kifonidis, K., Plewa, T., Scheck, L., Janka, H.-T., Muller, E. 2006, A&A, 453,
661
Kotake, K., Sato, K., & Takahashi, K. 2006, Rep. Prog. Phys., 69, 971
Kotake, K., Yamada, S., & Sato, K. 2003, ApJ, 595, 304
Laming, J. M. 2007, ApJ, 659, 1449
Marek, A., & Janka, H.-T. 2007, ApJ, submitted (arXiv:0708.3372)
Ohnishi, N., Kotake, K., & Yamada, S. 2006, ApJ, 641, 1018
. 2007, ApJ, 667, 375
Quirk, J. 1994, Int. J. Numer. Method. Fluids, 18, 555
Scheck, L., Plewa, T., Janka, H.-T., Kifonidis, K., Muller, E. 2004, Phys. Rev.
Lett., 92, 011103
Shen, H., Toki, H., Oyamatsu, K., & Sumiyoshi, K. 1998, Nucl. Phys. A, 637,
435
Stone, J. M., & Norman, M. L. 1992, ApJS, 80, 753
Wang, L., Howell, D. A., Hoich, P., & Wheeler, J. C. 2001, ApJ, 550, 1030
Wang, L., Wheeler, J. C., Li, Z., & Clocchiatti, A. 1996, ApJ, 467, 435
Wang, L., et al. 2002, ApJ, 579, 671
Woosley, S., & Janka, T. 2005, Nature Phys., 1, 147
Yamasaki, T., & Foglizzo, T. 2008, ApJ, in press (arXiv:0710.3041)
Fig. 16.Convergence tests. (a) Time evolutions of the normalized amplitudes jc
m
l
/c
0
0
j in the linear phase. The single-mode perturbation with l 1, jmj 1 is im-
posed initially. The models with 300 ; 30 ; 60 and 300 ; 60 ; 120 mess points are referred to as MESH0 and MESH1, respectively. (b) Time-averaged power spectra
jc
m
l
/c
0
0
j
2
. The average is taken over 150 t 400 ms. In this comparison, the randommulti-mode perturbation is imposed. [See the electronic edition of the Journal for a
color version of this qure.]
IWAKAMI ET AL. 1222